I am using a multi-processor machine, which has 4 processors and each processor has 8 cores. This machine also connects to many SSDs with 3 LSI cards. The problem I experience is that threads running on some processor can access SSDs faster than others. It seems some CPUs are closer to some SSDs than other CPUs. I'm not sure about the hardware hierarchy. Is it because a PCI bus is wired to connect to only one processor directly? so if another processor wants to access the devices on the PCI bus, it has to talk to the processor directly connecting to the bus first?
I wonder if anyone experiences the same thing. Do we have to use a CPU closer to the block devices to access them or is there a way of getting around it?
Could be a NUMA architecture - that certainly has memory (and cache) in regions that are tied to the node (may be several CPUs per node). What does this show ?
|All times are GMT -5. The time now is 10:46 PM.|